Skip to main content
Name one company and match every company that resembles it. The comparison uses a name-free description vector, so it keys on what a company does — not what it’s called, not what industry code it carries, not who it shares investors with.
This is the fastest path from “our best customers look like this” to a list, because you don’t have to articulate what they have in common. You point at one and we work it out.

​
Name the seed company

POST /v2/contact/company/enrich returns RBID_ORG for a domain, so you can resolve once and reuse the id across many searches.

​
Control how close a match must be

Two ways to do it. Pick one — sending both match and maxDistance is a 422.

​
Option 1 — a named preset

Send match with one of four names. Each is a similarity floor calibrated for look-alikes specifically. On target is the measured share of matches at that stop that genuinely fit. Presets are cumulative — broad returns everything related would, plus a looser band beneath it.

​
Option 2 — your own similarity threshold

Send maxDistance instead and put the floor wherever you like:
A 75% similarity floor is "maxDistance": 0.25. A raw maxDistance means exactly what it says and skips our calibration. To find your own number, preview with "includeScores": true and read _companySimilarity on each row — they come back best-match first. Scroll to where results stop being useful and set your floor just above that. The POST /v2/contact/company/lookalike endpoint also returns a similarity score per match, which is a quick way to see the shape of a seed’s neighbourhood before committing to a threshold.

​
If you send neither

You get similar — a 70% similarity floor. Sending nothing is exactly the same search as sending "match": "similar", down to the same rows. exact and similar are both strong. Past that, accuracy halves — related is a coin flip and broad is exploratory. Use the loose stops to find categories you hadn’t considered, then tighten.

​
Two endpoints, two jobs

Use the endpoint when you want to see and judge the neighbours, ranked by how close they are. Use the filter leg when you want a set cut at a threshold, combined with your other filters. Both take the same match / maxDistance settings above, and both default to similar. The endpoint additionally takes an optional, uncapped limit — it pages what the floor already selected, so omitting it returns every company above the floor. Its response echoes the match and minSimilarity it ran at, plus a per-match similarity score.

​
Any of several seeds

Up to ten look-alike nodes per filter — far more headroom than the other two targets. Join them with OR so a company qualifies by resembling any seed:
Put look-alikes in an AND group and a company must resemble every seed at once. That combination almost never matches anything. Use OR unless you specifically want the intersection.
Feeding in your top ten closed-won accounts is the highest-value version of this: it describes your ICP from evidence rather than from a positioning document.

​
Exclude a look-alike

There is no “not in” look-alike node. Wrap it in a group and negate:
Useful for carving a segment out of a broad sweep — everyone like our customers, minus anyone like the incumbent we never beat.

​
When a company has no vector

A company with no usable description has nothing to compare against. The 404 is deliberate: it lets you tell “no similar companies exist” apart from “we can’t assess this company”. If a seed returns 404, pick a different one — usually a larger company in the same space.

​
Next steps